隨著近年來Google Map, Yelp, Foursquare等適地性社群網路的蓬勃發展，對於精準的興趣點推薦系統的需求也變得越發重要，然而適地性社群網路因地理距離限制而會產生資料稀疏等問題。為了緩解資料稀疏的影響，可以在基於協同過濾的推薦方法中加入其他輔助資訊，如物品的描述文件或是使用者個人資訊等。
本研究主要目的為研發針對興趣點特徵考量之推薦系統，我們加入了地理因素以及深度學習使用者評論之特徵並與基於模型之推薦算法矩陣分解同時訓練，藉此在推薦時能考量評論語義以及地理關係之特徵，得到比起傳統推薦方法還要更高的推薦效能。
為了評估本研究提出方法的效能，我們使用兩種不同地理分布的適地性社群網路資料集，分別為Google Map與Yelp資料集，並利用兩種不同的評測推薦系統方法來比較本研究與其他研究之模型。結果顯示我們提出的方法能在兩種不同的資料集上都取得有效的提升，歸因於我們提出的方法能夠同時考量興趣點中的地理特徵以及從使用者評論取得的描述特徵。

With the increase of data in LBSN (Location-based Social Network) such as Google Map, Yelp, Foursquare, etc. The demand of accurate POI recommendation system has become more critical. However, LBSN also suffer from severe user-item data sparsity problem due to the geographical limitation. To alleviate this problem, collaborative filtering (CF)-based approaches can utilize additional information such as item description document, user profile or social network to model the feature of user and item. In this study, we aim to build a POI recommendation system specifically focused on the characteristics of POI. That is, we take the geographical influence and user review semantic information into consider and jointly combine these features into model-based approach matrix factorization for model training. By doing so, our model can benefit from both semantic features and geographical information and get better recommendation performance than traditional recommendation system methods. To evaluate the overall performance of our model, we conduct several experiments on two different real-world LBSN datasets Google Map and Yelp respectively. The result shows that our model outperforms all other baselines in two different geographical distribution datasets. Our method makes it possible to consider geographic and semantic information simultaneously with matrix factorization and further enhance the prediction ability.

[1] G.Adomavicius and A.Tuzhilin, “Toward the next generation of recommender systems: A survey of the state-of-the-art and possible extensions,” IEEE Trans. Knowl. Data Eng., vol. 17, no. 6, pp. 734–749, 2005.
[2] J. L.Herlocker, J. A.Konstan, L. G.Terveen, and J. T.Riedl, “Evaluating collaborative filtering recommender systems,” ACM Trans. Inf. Syst., vol. 22, no. 1, pp. 5–53, 2004.
[3] W.Wang, “Point of Interests Recommendation in Location-Based Social Networks,” p. 109, 2017.
[4] G.Ling, M. R.Lyu, and I.King, “Ratings meet reviews, a combined approach to recommend,” Proc. 8th ACM Conf. Recomm. Syst. - RecSys ’14, pp. 105–112, 2014.
[5] S.Li, J.Kawale, and Y.Fu, “Deep Collaborative Filtering via Marginalized Denoising Auto-encoder,” Proc. 24th ACM Int. Conf. Inf. Knowl. Manag. - CIKM ’15, pp. 811–820, 2015.
[6] H.Wang, N.Wang, and D.-Y.Yeung, “Collaborative Deep Learning for Recommender Systems,” 2014.
[7] S.Purushotham, Y.Liu, and C.-C. J.Kuo, “Collaborative Topic Regression with Social Matrix Factorization for Recommendation Systems,” 2012.
[8] Y.Koren, “Factorization meets the neighborhood: a Multifaceted Collaborative Filtering Model,” Proceeding 14th ACM SIGKDD Int. Conf. Knowl. Discov. data Min. - KDD 08, p. 426, 2008.
[9] Y.Hu, C.Volinsky, and Y.Koren, “Collaborative filtering for implicit feedback datasets,” Proc. - IEEE Int. Conf. Data Mining, ICDM, pp. 263–272, 2008.
[10] K. W.Church and P.Hanks, “Word association norms, mutual information, and lexicography,” Proc. 27th Annu. Meet. Assoc. Comput. Linguist. -, vol. 16, no. 1, pp. 76–83, 1989.
[11] P. D.Turney, “Mining the Web for synonyms: PMI-IR versus LSA on TOEFL,” Proc. 12th Eur. Conf. Mach. Learn. (ECML-2001), Freiburg, Ger., pp. 491–502, 2001.
[12] P. D.Turney and P.Pantel, “From Frequency to Meaning:Vector Space Models of Semantics,” J. Artif. Intell. Res., vol. 37, pp. 141–188, 2010.
[13] O.Levy and Y.Goldberg, “Neural Word Embedding as Implicit Matrix Factorization,” Adv. Neural Inf. Process. Syst., pp. 2177–2185, 2014.
[14] D.Liang, J.Altosaar, L.Charlin, and D. M.Blei, “Factorization Meets the Item Embedding: Regularizing Matrix Factorization with Item Co-occurrence,” Proc. 10th ACM Conf. Recomm. Syst. - RecSys ’16, pp. 59–66, 2016.
[15] Y.LeCun, L.Bottou, Y.Bengio, and P.Haffner, “Gradient-based learning applied to document recognition,” Proc. IEEE, vol. 86, no. 11, pp. 2278–2323, 1998.
[16] A.Severyn and A.Moschitti, “Learning to Rank Short Text Pairs with Convolutional Deep Neural Networks,” Proc. 38th Int. ACM SIGIR Conf. Res. Dev. Inf. Retr. - SIGIR ’15, pp. 373–382, 2015.
[17] D.Kim, C.Park, J.Oh, S.Lee, and H.Yu, “Convolutional Matrix Factorization for Document Context-Aware Recommendation,” Proc. 10th ACM Conf. Recomm. Syst. - RecSys ’16, pp. 233–240, 2016.
[18] Y.Kim, “Convolutional Neural Networks for Sentence Classification,” pp. 1746–1751, 2014.
[19] N.Kalchbrenner, E.Grefenstette, and P.Blunsom, “A Convolutional Neural Network for Modelling Sentences,” Proc. ACL, pp. 655–665, 2014.
[20] W. R.Tobler, “A Computer Movie Simulation Urban Growth in Detroit Region,” Econ. Geogr., vol. 46, pp. 234–240, 1970.
[21] Y.Yu and X.Chen, “A Survey of Point-of-interest Recommendation in Location-based Social Networks,” pp. 53–60, 2015.
[22] T.Mikolov, K.Chen, G.Corrado, and J.Dean, “Distributed Representations of Words and Phrases and Their Compositionality,” pp. 1–9.
[23] J.Pennington, R.Socher, and C. D.Manning, “GloVe : Global Vectors for Word Representation.”
[24] G.Adomavicius and J.Zhang, “Impact of data characteristics on recommender systems performance,” ACM Trans. Manag. Inf. Syst., vol. 3, no. 1, p. 3:1-3:17, 2012.
[25] C.-J.Lin, “Projected Gradient Methods for Nonnegative Matrix Factorization,” Neural Comput., vol. 19, no. 10, pp. 2756–2779, 2007.